Career & Practice
Top AI Research PhD Programs: A Researcher's Guide
Twenty programs across the US, UK, Europe, Canada, and Asia — real program structure, admission statistics where they're actually public (most aren't), faculty links, and funding, verified through direct research rather than copied from a rankings list. Where a number couldn't be confirmed, this article says so instead of guessing.
Before you rely on the numbers below: current-year acceptance rates are, with only a couple of exceptions, simply not published by these departments — most searches for "Stanford CS PhD acceptance rate" or similar return old, third-party-estimated figures, not official current data. Every fact below carries a HIGH / MEDIUM tag: HIGH means confirmed on the university's own official page; MEDIUM means found only via a secondary source or a search-result summary of an official page, not a direct fetch. Where nothing was found, this article says "not publicly disclosed" rather than estimating a plausible-sounding percentage.
Why "Top" Is Contested
CSRankings, Briefly
United States
UK & Europe
Canada
Asia
Admissions Reality Check
Program vs. Advisor
Confidence Notes & Gaps
Full Reference List
Why "Top" Is Contested
Ask for a ranking of AI PhD programs and you'll get a different answer depending on whether the source counts publications (CSRankings.org), surveys department reputations (US News), or just repeats whichever list a previous article cited. This piece skips the single-number ranking entirely and instead reports what's actually verifiable per program: how the degree is structured, whether admission numbers are public, who the current faculty are, and how funding works — the things you'd actually need to know to apply, not a league-table position.
CSRankings, Briefly
CSRankings.org is purely publication-metrics-based: it counts faculty publications appearing in a curated list of the most selective conferences per subfield, with each paper's credit split among its faculty co-authors only — student co-authors don't dilute or add institutional credit the way they would on a naive per-author count. This is explicitly designed to be harder to game than a reputation survey or a raw citation count, though the exact credit-splitting mechanics are worth checking directly against the CSRankings GitHub repository before quoting a precise formula. It's a genuinely different measurement than US News' reputation-survey approach, and the two rankings disagree meaningfully — worth knowing before treating either as "the" ranking.
United States
| Program | Structure | Admission Stats | Faculty / Funding |
|---|---|---|---|
| Stanford (SAIL) | AI is a concentration within the general Stanford CS PhD, not a separate degree HIGH | Not publicly disclosed MEDIUM | Faculty: ai.stanford.edu/faculty. GRE not required; funded stipend+tuition standard HIGH |
| MIT (CSAIL) | CSAIL is a lab; students are admitted via EECS HIGH | ~4,400 applications for the 2025 cycle, EECS-wide HIGH | Faculty directory: csail.mit.edu/research/all-groups. Standardly funded MEDIUM |
| CMU | A genuinely separate, dedicated Machine Learning Department PhD, distinct from the general CS PhD HIGH | Not publicly disclosed for the PhD (a commonly-cited "4%" figure is for CMU's Master's in ML, not the PhD — don't conflate) MEDIUM | J. Zico Kolter (department head) — faculty page. GRE optional and explicitly not a disadvantage; no GPA cutoff used HIGH |
| UC Berkeley (BAIR) | BAIR is a lab; degree is the EECS PhD HIGH | Not publicly disclosed currently; only dated (2011/2018) third-party figures around 3% exist and shouldn't be treated as current MEDIUM | Faculty directory: bair.berkeley.edu/faculty. Standardly funded MEDIUM |
| Princeton | AI is one of three core breadth areas (with Systems and Theory) in the general CS PhD HIGH | Official annual metrics exist at program-metrics, but current CS-specific numbers weren't extracted here — check the live table MEDIUM | ~15% of Princeton grad students university-wide get full fellowships, ~85% get some aid (not CS-specific) MEDIUM |
| U. Washington (Allen School) | General Allen School PhD HIGH | Over 3,000 applications in a recent cycle, average admit GPA ~3.8 MEDIUM | GRE "no longer required or accepted... will not be reviewed even if submitted" HIGH |
| Cornell | General CS PhD, competency required across four core areas including AI HIGH | Not publicly disclosed for CS specifically MEDIUM | Standard fully-funded CS PhD assumed, not explicitly confirmed MEDIUM |
| UT Austin | General CS PhD, ML/DL/NLP among research areas HIGH | Not publicly disclosed for the PhD (undergraduate CS admit rates ~5-8% exist but are a different, unrelated program) MEDIUM | Standard fully-funded CS PhD assumed MEDIUM |
UK & Europe
| Program | Structure | Admission Stats | Faculty / Funding |
|---|---|---|---|
| Oxford | DPhil in Computer Science; AI-specific funded routes via CDTs (e.g. Fundamentals of AI, AIMS) HIGH | Official divisional admissions stats exist, but no single clean DPhil acceptance rate HIGH that data exists / not found for one clean number | Home-fee DPhil applicants automatically considered for funding; named schemes include Oxford–Google DeepMind Graduate Scholarships and Clarendon Scholarships HIGH |
| Cambridge (MLG) | The well-known Machine Learning Group PhD track is administratively in the Department of Engineering, not Computer Science — a genuine structural nuance HIGH | Not publicly disclosed; the group reportedly can't admit many qualified, even funded, candidates due to volume MEDIUM | Zoubin Ghahramani, Carl Edward Rasmussen, Richard Turner, José Miguel Hernández-Lobato, Adrian Weller among MLG faculty — mlg.eng.cam.ac.uk MEDIUM |
| UCL (Gatsby Unit) | Distinctive 4-year MPhil/PhD combining ML and computational neuroscience — not a standard CS PhD HIGH | Not publicly disclosed | "Full funding is available regardless of nationality" — stated explicitly on the official FAQ HIGH |
| ETH Zurich | PhD via individual departments; AI coordinated cross-departmentally via the ETH AI Center (140+ professorships across 16 departments), a funding/affiliation mechanism not a separate degree HIGH | Professor-sponsorship-first model; a single admit-rate number isn't very meaningful here MEDIUM | Doctoral students are standardly salaried employees (Swiss university norm) MEDIUM |
| EPFL (EDIC) | EDIC — a unified CS doctoral program covering AI, theory, systems, ~80 affiliated faculty HIGH | Described only as "very competitive," no numeric rate found | "Financial support in the form of a salary is available to all admitted PhD students" — stated explicitly HIGH |
| MPI-IS (Tübingen) | Doctoral students enroll via IMPRS-IS, a partnership with Tübingen and Stuttgart — no standalone MPI degree; also a distinct Cambridge–Tübingen Fellowship and the Max Planck ETH Center for Learning Systems HIGH | Not publicly disclosed; institute-wide "100+ faculty, 400 enrolled students" cited MEDIUM | Standard German Max Planck doctoral positions are salaried, not stipend-based MEDIUM |
| U. Amsterdam (AMLab) | Dutch model: Master's first, then employed as a PhD researcher within AMLab HIGH | Not publicly disclosed | Max Welling, Jan-Willem van de Meent (director), Herke van Hoof among AMLab faculty MEDIUM. Dutch PhDs are standardly salaried employee contracts MEDIUM |
Canada
| Program | Structure | Admission Stats | Faculty / Funding |
|---|---|---|---|
| U. Toronto | General CS PhD; many ML faculty are also Vector Institute affiliates/CIFAR AI Chairs — an affiliation model, not a separate degree HIGH | Explicitly not publicly disclosed, per a source noting Toronto is among the programs that don't release this data MEDIUM | Vector Institute: ~143 faculty/affiliates (38 CIFAR AI Chairs) as of 2023 MEDIUM |
| Mila / U. Montréal | Students register through their supervisor's home university (UdeM's DIRO, McGill, Polytechnique, or HEC Montréal) — Mila is the research umbrella, not the degree-granting body; a supervisor pre-agreement is required before applying HIGH | Not publicly disclosed | Typically funded via supervisor grants/institute funding MEDIUM |
| U. Alberta (Amii) | General Computing Science PhD, RL research concentrated via the Amii affiliation HIGH | Not publicly disclosed; described qualitatively as very competitive MEDIUM | Martha White — Canada Research Chair in RL, faculty page HIGH. RL-lab students typically fully funded as RAs MEDIUM |
Asia
| Program | Structure | Admission Stats | Faculty / Funding |
|---|---|---|---|
| Tsinghua (College of AI) | Dedicated College of AI; first-year PhDs rotate 3–6 months through multiple supervisors' groups before settling HIGH | Planned to admit 50 PhD students in 2025, per the official page (a target, not necessarily final admits) HIGH | Andrew Chi-Chih Yao (Turing Award laureate) leads the College HIGH |
| Peking University | General CS PhD, open to master's-holders and some direct bachelor's-to-PhD applicants HIGH | Not publicly disclosed | 109 faculty (56 professor/researcher-level, incl. 7 Academicians, 4 Chair Professors). "All students admitted... will receive full financial support, including tuition and monthly stipend" — stated explicitly HIGH |
| NUS (School of Computing) | General PhD (by Research), AI as one research area among several HIGH | Not publicly disclosed | Research Scholarships: monthly stipend + full tuition subsidy, renewable up to 4 years, plus a S$500/month top-up after passing the qualifying exam — stated explicitly HIGH |
| KAIST | Advisor-first model: applicants must name one specific faculty member as desired advisor at application time, including for the Kim Jaechul Graduate School of AI HIGH | Not publicly disclosed | Tuition waiver + monthly stipend standard; College-of-Engineering-wide average stipend cited over 1.7M KRW/month (not AI-School-specific) MEDIUM |
Admissions Reality Check
1
GRE is widely dropped, not universally. UW's Allen School will not even review a submitted GRE score. CMU treats it as optional and explicitly not a disadvantage. Stanford's CS PhD doesn't require it either — though other Stanford schools, like the GSB, still do, so don't assume a university-wide policy from one department's page.
2
Research fit is stated to outweigh GPA at the programs that say anything explicit about it. CMU states plainly that faculty research-interest alignment is "a very significant part of the admissions decision" and that it does not rank applicants by GPA or use a cutoff. Stanford's own checklist advises naming two or three specific faculty whose work matches your interests, rather than describing broad interest in "AI."
3
Most acceptance-rate numbers you'll find online are stale or unofficial. Real, current, department-disclosed numbers exist for exactly two programs in this list — MIT EECS's ~4,400 applicants and Tsinghua's planned 50-student target. Everything else either isn't published at all, or the number circulating online is several years old (Berkeley's often-cited figures trace to 2011 and 2018).
4
Advisor-first models exist and change how you apply. KAIST requires naming a specific desired advisor at application time. Mila requires a supervisor pre-agreement before you can apply at all — the research institute itself doesn't grant the degree; your eventual advisor's home university (UdeM, McGill, Polytechnique, or HEC Montréal) does.
Program vs. Advisor
Every admissions page that says anything specific about what it's looking for says some version of the same thing: fit with a specific lab matters more than the institution's overall reputation. That's consistent with what the study stack article and the AI Researcher Atlas both point at from different angles — the person you'd actually work with under matters more than the name on the diploma.
A practical checklist for evaluating a specific advisor, not just a program: Is their publication output recent and active, not a few strong papers from years ago? Is their funding visibly stable (grants, named fellowships, industry partnerships) rather than dependent on one expiring source? What happened to their recent graduated students — did they land research roles at labs or in academia, or mostly leave the field? Advisor fit checked this way tells you more about your actual next five years than any program-wide ranking does.
⚠️ Confidence Notes and Gaps
Program structure facts (dedicated department vs. concentration within a general CS PhD, degree-granting mechanics, advisor-first admissions) are consistently high confidence across this list, since universities describe their own structure clearly on official pages. Admission statistics are the opposite: nearly every program in this list does not publicly disclose current acceptance rates, and the few numbers that do circulate online are frequently outdated (Berkeley) or refer to a different program entirely (CMU's often-cited "4%" is the Master's, not the PhD). Some facts here — including several UK/European faculty names and a few CMU official-page details — were confirmed through search-result summaries of official pages rather than a direct fetch of the page itself, and are flagged MEDIUM accordingly; verify directly before citing them formally. If you're actually applying, treat every admission number in this article as a starting point for your own search, not a final answer — check the live official page for the current cycle.
🔗 Full Reference List
Stanford CS PhD admissions
MIT EECS admission process
CMU doctoral admissions FAQ
UC Berkeley EECS admissions
Princeton Graduate School program metrics
UW Allen School PhD admissions FAQ
Oxford DPhil in Computer Science
Cambridge Machine Learning Group PhD
UCL Gatsby Unit PhD programme
ETH AI Center PhD fellowships
EPFL EDIC doctoral program
IMPRS for Intelligent Systems (MPI-IS)
Mila PhD admission process
Amii — Reinforcement Learning at Amii
Tsinghua College of AI graduate programs
Peking University CS PhD (international admissions)
NUS Computing PhD scholarships
KAIST doctoral admission
CSRankings.org methodology (GitHub)
A Public Study Stack for Frontier AI Research Roles (companion article)
The AI Researcher Atlas (companion article)